an implementation of real and accurate cloud based … · implementation of real and accurate cloud...

http://www.iaeme.com/IJARET/index.asp 51 [email protected]

International Journal of Advanced Research in Engineering and Technology (IJARET) Volume 11, Issue 10, October 2020, pp.51-66, Article ID: IJARET_11_10_006 Available online at http://www.iaeme.com/IJARET/issues.asp?JType=IJARET&VType=11&IType=10

ISSN Print: 0976-6480 and ISSN Online: 0976-6499

DOI: 10.34218/IJARET.11.10.2020.006

© IAEME Publication Scopus Indexed

AN IMPLEMENTATION OF REAL AND

ACCURATE CLOUD BASED FRAUD

DETECTION SYSTEM BY USING DEEP AND

MACHINE LEARNING TECHNIQUES

Prasanthi Gottumukkala

Department of Information Technology CMR Engineering

College, Hyderabad, India.

Dr .G. Srinivasa Rao

Department of Information Technology, GITAM

University, Visakhapatnam, India.

ABSTRACT

From the last decades deep and machine learning techniques are achieves

prominent outcomes in various areas like data analytics, big data processing and cloud

classifications. Which makes the real time intelligent models, in this investigation

proposes a cloud based fraud detection system by using Fully Convolution Neural

Networks (FCNN). Along this deep learning, Gradient Boosting Machine Learning

(GBML) technology is incorporated, it permits the real time frauds classification and

regression on clouds. Our proposed methodology FCNN and GBML compete with

existing models in terms of accuracy, recall, true positive rate and precision.

Keywords: Cloud Frauds, GBML, FCNN, Credit Card Frauds.

Cite this Article: Prasanthi Gottumukkala and Dr .G. Srinivasa Rao, An

Implementation of Real and Accurate Cloud Based Fraud Detection System by Using

Deep and Machine Learning Techniques, International Journal of Advanced Research

in Engineering and Technology (IJARET), 11(10), 2020, pp.51-66

http://www.iaeme.com/IJARET/issues.asp?JType=IJARET&VType=11&IType=10

1. INTRODUCTION

The number of financial transactions via credit card increases drastically and with it various

frauds has occurred. From 2019 financial association and professionals survey the payment

frauds observed clearly. This survey of records analyzes the frauds around 80% of all

organizations. As rise of online payments billions of frauds can occur on clouds. Because of

this reason banks and financial organizations challenge to construct an effective fraud detection

system. Deep learning presenting a many applications with active solutions, this can helps the

management traders, risk analyzers and fraud detectors. In digital landscape, for example, deep

learning is used to build an intelligence software that can directly interact with people and gives

An Implementation of Real and Accurate Cloud Based Fraud Detection System by Using Deep

and Machine Learning Techniques


quires. In trading, trade analyze system is utilized to made fast decisions. More over the general

usage of deep learning in banking sectors is the security against frauds. With the help of deep

learning and machine learning techniques identify the suspicious fraud activities with an easy

way.

Depending upon ID’s of transaction history, deep learning shows the behaviors of

customers and machine learning detects fraudster data with efficient classification, if there is a

fraud or not. In this research, the main objective is cloud based frauds detection system design

using modern techniques.

Deep learning presents a very significant solution to deal with fraud transactions on clouds,

making to accurate use of “financial organizations-big data”. In deep learning many methods

are there, in this fully convolution neural networks (FCNN) are utilized for preprocessing stage.

At final stage, classification purpose gradient boosting machine learning technique is to be used.

FCNN currently has given that accurate precisions and throughputs to many problems.

In this data investigation, clouds related fraud detection with binary categorization trouble

are investigated, this FCNN predict the binary fields with efficient manner. So clouds-frauds

can easily identified by the FCNN and classify the “legitimate” or “fraudulent”. The binary

classification is an easiest classification, where a compilation of information is classified into 2

classes, depending upon features available.

Based upon transaction history machine learning optimization technique with new methods

analyze the fraud detection ratios and identifies the transaction is fraud or not [15]. MCKNSAY

[5] explains that an important solution to deal the frauds at big data related heavy transactional

traffic. In FCNN multi-level pooling layers and hidden layers learning model is discussed in

[7].

The best precision and efficiency generating system promising the results in many fields at

binary classification. In big data analysis cloud based fraud detection system with binary

classification is analyzed and classified for specific outcomes. There are many numerical

models for binary learning classifications such as DT- Decision trees, NN- Neural networks,

SVM- Support vector machine, KNN- K-nearest neighbors and logistic regression [8].

Figure 1 Fraud prevention system.

The real time information which has been collected from john a stanko dataset, this dataset

consist of various control actions on receiving data, processing data and returning data at the

time of transaction[11]. Our proposed model is a novel work, it is useful for real time

applications at the time of cloud based fraud classification. The deep learning networks utilizing

the auto encoder, classifies the transactions and real time decision, resulting that fraudulent

Prasanthi Gottumukkala and Dr .G. Srinivasa Rao


identification data is the output. Gradient boosting machine learning model is used for testing

and classification. Along this, support vector machines are also used for binary classification

of cloud based frauds but not efficient [17]. The present work performed based on gradient

boosting machine learning model and compare with SVM, LR and KNN etc.[6]. The SVM’s

drawn a major attention in current studies because of its major performance depending on

classification action. However in some of conditions ANN had given efficient results compare

with nonlinear models [18]. As for datasets one example exercise is performed on European

cards transaction history. This work gives two key points, which are illustrated below.

A real time full convolution neural networks (FCNN) approach on credit card fraud

detection problem for clouds based on auto encoder.

Comparison of different classification methods for observation of auto encoder

functionality.

This discussion and implemented design compete with current technology and achieves

better improvement. The rest of the work is organized as like in section 2 discuss a short review

on related methods. In section 3 describes FCNN architecture and GBML classification. In

section 4 explains testing environment, then analyze the test out comes. Finally in section 5 we

gives a future work and conclusion.

2. RELATED METHODS

From few years back many alternate solutions are proposed for clouds fraud detection system.

The significant methods are statistical or artificial intelligence [10] has been used based on

illustrated examples such as

Regression model

SVM

Multiple discriminant analysis [23].

2.1. Literature Survey

The logistic regression optimization method utilized for binary classification [24], compare

various modern models for fraud detection [15] improves the accuracy. Like various methods

were used for cloud based problem detection system. The brief discussion on various multi

model features explained below table 1 clearly.




Table 1 literature survey

ARTICLE NO SURVEY ARTICLE

METHODOLOGY

OF

COMMUNICATION

TECHNIQUE KEY POINTS

[1] Fraud Control Machine learning EC 2 layer problems

State of art framework

Cloud security

[2] Global payments group Data mining encoder

approach

Fraud identification

Fraud classification

Real time database analysis

[3] Business data mining Machine learning Classification

Regression

Analysis

[4] Binary classification Big data and cloud

computing

classification

Financial market analysis

Cloud computing techniques

[5] Fraud analysis Quick estimation

Account Take over

method

Knowledge based

authentication

Fraud landscape shifting

EVM security on clouds

ARTICLE NO SURVEY ARTICLE

METHODOLOGY OF

COMMUNICATION

TECHNIQUE KEY POINTS

[6] Statistical theory Machine learning Productive function

statistical learning algorithm

[7] MIT press Deep learning Cloud data fraud analysis

EC2 layer estimation

efficiency, accuracy

[8] Survey of data mining Deep learning Tensor flow calculation

Accuracy, True positive rate

[9] Statistical fraud

detection

Statistics and machine

learning effective

technologies data

analytics method.

Fraud detection

Digital cloud audit

mechanism.

Transactions history analysis.

[10] Data mining related

fraud detection

Rank based fraud

detection

Matec web content dataset

Credit score

Fraud or not?

[11] Real time computing MISCONCEPTION Cloud efficiency EC2 server

Sensitivity

[12] Credit card frauds Conventional SVM

method

Fraud identification

Server boosting real time

efficiency

[13] Credit card computer

system

Auto encoder Cloud based frauds analysis

EC2 cloud fraud

identification

[14] Fraud detection

technique

Analysis of credit card

fraud using deep

learning

Efficiency

Accuracy

Throughput

[15] Data mining for credit

card

Neural networks Credit card fraud detection

Linear discriminant analysis

Recall and F1 score



The mobile computing and cloud computing technologies are facing many issues at the time

of business transactions. The machine learning models with random forest optimization solves

the many issues in the cloud computing. But, this is a traditional search process, therefore the

delay of operations takes more time [26]. The mechanism of business analysis and service

possible with cloud and mobile computing technologies. Such that an advanced cloud

computing and security tools are necessary to analyze the frauds. The block chain and big data

technologies mainly facing security issues, without data securities these technologies are going

to be affordances. The decision making and application computing May takes more time with

less accuracy. These type of partial results degrade the cloud usage. The data computing and

cloud security management is a major driven approach at data mining. Now a days the major

IT business depending on cloud based applications, if this services are cannot give the assurance

then automatically fraudsters are attacking cloud applications. The theoretical and experimental

models deal the current issues of cloud security and gives the better decision making [27]. The

data security and network issues are major challenges in cloud system, if fraudsters hack clouds

then applications getting trouble. Therefore an advanced technology is necessary to implement

the security to clouds [28]. With exponential growth in IoT and edge computing technologies

helps the clouds and its applications. The generalized cloud environment can deals the moderate

security issues, but real time monitoring is necessary. For these applications purpose we can

design the accurate cloud security model with low latency time [29].

In this study [30] cloud computing and its services are explained clearly, the layers security

and cryptography protection concepts are discussed. Various cloud services and related security

issues managed by computing features concept. The challenges like security, computing and

platform economy features are major roles in cloud computing technology.

3. METHODOLOGY

3.1. Prediction model-FCNN

In this section our proposed classification method using FCNN is designed based on pooling

layer and hidden layer.

Figure 2 FCNN for fraud detection.

A periodical online credit card fraud training dataset is given to input of our design and predict

the frauds based on FCNN mathematical computations. The deep learning is a user friendly

mechanism it can help any type of platforms like clouds, data mining and big data analytics.

FCNN network can normalize the dropouts by using rectified linear units (ReLU) function. This

function helps for various layers in the FCNN network, in all convolutional network the final

layer is 3*3 sigmoid activation layer.




Figure 3 FCNN layer analysis.

This mechanism very helpful for cloud based dropdown network issues at every time, all

FCNN filters searching the ReLU activated sub layers for fraud detection. In this N=56 FCNN

CON 2 dataset is used to identify the computed frauds in the clouds, which is shown in fig2

clearly. Fig 3 describes about general architecture of layer information which can operated

using 300*300*56 confuse matrix. More than 500 trained weights are predicted using FCNN

keras library, FCNN processed the ReLU activation function and convolute the filter layers

which is shown in above figure.

Figure 4 FCNN complete architecture.



The FCNN is process the normalized ReLU functions using 32 to 64 CNN layers, our design

classified into three residuals which are illustrated below. 1. A regular depth of cloud computing

key parameters and normalized dropout layers has described. 2. Replace the last convolutional

layer with altered layers such as 1, 2, 4, 8, 16 etc.. 3. The entire architecture is combination of

second and 3rd layers. Using FCNN CON 2 dataset is input to our training model and validate

the target test data.

Figure 5 Auto encoder method.

We also compare the FCNN with CNN, results that FCNN residuals got best performance.

In this FCNN deep learning is used as auto encoder, it is a neural network and equalize the input

and output functions. Which is shown in above figure4.

Fig 5 explains that auto encoder and decoder model, which consist of two parts 1. Encoder:

Identify the more number of frames at maximum point operations. At this point clouds may be

loose their security issues, therefore chance of frauds may occur on original inputs. 2. Decoder:

The input data is able to decode with reconstructed data and it is fed back to encoder stage. The

entire operations are shown in below equations 1to 3.

Figure 6 average convolution prediction.

We can write the same for all m neurons in layer

𝑍𝑖[𝑙]

= 𝑊𝑖𝑇 . a [l-1] + bi 𝑎𝑖

[𝑙] = g[I] (𝑍𝑖

[𝑙]) (1)

Fig 6 explains that tangent function analysis at automatic convolution and data

computations. The fraud information is predicted easily by using the average nature of gradient

FCNN realization.




Figure: 7. Deep Learning analysis.

𝑍1[1]

= 𝑊1𝑇 . a[0] + b1 𝑎1

[1] = g[1] (𝑍1

[1]) (2)

𝑍2[2]

= 𝑊2𝑇 . a[1] + b2 𝑎2

[2] = g[2] (𝑍2

[2]) (3)

Equation 1 to 3 are major equations, these are used to find out the Impulse response of first

connected layer in Convolutional neural network. Here Z is impulse function, W is weight of

particular tree, a & b are co-expression knowledge in FCNN. Fig 7 explains that over all deep

learning analysis of FCNN model, this model can implement on python 3.6, keras GPU, GPX

660, Memory of 16GB. In this y represents of output of cloud based fraud count. In this overall

information of fraud has detailed like date, time, location and platform. By the help of following

equations which are shown in above equations 4

𝑍𝑛[𝑁]

= 𝑊3𝑇 . a[n] + b3 𝑎3

[𝑛] = g[n] (𝑍3

[𝑛]) (4)

RELU (x) = {0 𝑖𝑓 𝑥 < 0

𝑥 𝑖𝑓 𝑥 > = 0 (5)

The above equation 5 is the final FCNN output which gives the rank of prediction, this rank

is used to predict the frauds in the cloud system. For classification we are moving to GBML

technique, which is explained in below section. The 1st layer in FCNN is connected with 81

neurons, it is getting information from all layers in cloud. The 2nd layer is convolutional layer

it is designed with 20 CNN neurons with 4x4 kernel size. The output of individual layers are

connected between kernel matrix and fitting matrix. The over fitting layers and internal

functions are managed by 3rd module in FCNN. The max pooling layers are interact with 4x4

size kernel and connected with last layer.

3.2. GBML-classifier

Gradient boosting is a powerful and popular machine learning technique, it is used in massive

big data analysis, cloud computing and data mining. GBML is a method which converts the

week classification model into strong classifier.



Figure 8 GBML for cloud frauds classification.

In boosting every tree is modified with validation of original dataset and can easily assigned

the weights to each tree. After evaluation of 1st tree can increases the weights of another

observations of trees. In this lower weights are very simple to classify the prediction, therefore

this GBML is tree1+ tree2...+tree N of residuals are classified. GBML trains the every datasets

which are taken from input and classify the frauds in the clouds at real time. Our main agenda

is to classify the credit faults compared to good transactions. This entire discussion is explained

by using the below mathematical computations.

3.2.1. Mathematical computations of GBML [15]

𝑟2𝑚 = −𝜕𝐿

𝜕𝑓(𝑥𝑖) (yi, f(xi)) (6)

Here r2m is the differential rank function which can differentiate the fitness function as per

Trees in gradient boosting mechanism.

𝛼 = argmin𝛼L (yi,𝑓𝑚−1 (x) + 𝛼ℎ𝑚(x)) (7)

Here alpha is the argument function, in this L is linear cross validation function. It is

differentiate the over fitting activity of weights.

Figure 9 Tree structure in GBML [15].

Loss = ∑ 𝐿(𝑦𝑖, 𝑓𝑚−1 (𝑥𝑖) + ℎ𝑚 (𝑥𝑖) ) 𝑛𝑖=1 + ∑ 𝛺(ℎ𝑚 (𝑥𝑗))𝑇

𝑗=1 (8)

After cross validation we want to estimate the loss function, here x, y are input

parameters, 𝛺 is fraud margin function. The selected dataset is estimating through

following equation 8.




Loss=∑ [𝐿(𝑦𝑖, 𝑓𝑚−1(𝑥𝑖)) +

𝜕𝐿

𝜕𝑓𝑚−1𝑥𝑖 (𝑦𝑖, 𝑓𝑚−1(𝑥𝑖))ℎ𝑚(𝑥𝑖) +

1

2

𝜕2𝐿

𝜕𝑓𝑚−1

(𝑥𝑖)2 (𝑦𝑖, 𝑓𝑚−1(𝑥𝑖))ℎ𝑚

2 (𝑥𝑖)

.]𝑛

𝑖=1 +

∑ 𝛺(ℎ𝑚 (𝑥𝑗))𝑇𝑗=1 Ω(hmT) (9)

Fig 9 & equation 9 is the GBML model training computations, these are explains about

cloud interaction depth and arguments. The cloud is a time varying platform, when it moves to

busy stage there may be a chance of frauds. This can handle the GBML method to prevent the

busy stage of clouds using loss function, gain function equations.

Figure 10 Tree classifier

Loss = −1

2∑ [

𝐺𝑗2

𝑥𝑗+ 𝜆]𝑇

𝑗=1 + ET (10)

Gain = Loss before – Loss a filter split

= - E + 1

2[

𝐺𝐿2

(𝐻𝑅+𝜆)+

𝐺𝑅2

(𝐻𝑅+𝜆)−

(𝐺𝐿+𝐺𝑅)2

𝐻𝑅+𝐻𝐿+𝜆] (11)

Fig 10 explains about internal calculations with ET (exact time) and variable interactions

validation, in this GBM assigns the number of trees and respective weights using this point

predictor classify the influence of cloud parameters.

4. EXPERIMENTAL RESULTS

4.1. Dataset

DN- CON2 dataset is used in this work, this is input of both FCNN predictor and GBML

classifier. The ULB group on big data and fraud detection datasets are available freely named

as Kaggle, which is shown in table 1 clearly.

Table 2 Dataset features

1. Data set Multivariate

2. Attributes Categorical, Integer

3. Associated function Classification

4. N.of instances 284997

5. N.of Attributes 40

6. Missing N/A



Table: 1 explains about features of DN-CON2 consists of various instances like 285000

around, type of attributes are either by an integers or by categorical model, and in this datasets

no missed data is placed.

Figure 11 Data set distribution.

Time and over sampled fraud classes count is shown in fig 11, which consist of 3.12% all

transactions represented by A0, A2, A3, ………………. An, above analysis is clearly explains

about no of frauds and regular transactions.

Example:

The above dataset consists of 284807 rows and 31 columns. The column number 1 to 28

are the principal FCNN + GBML component. The main reason to attain these frauds, only one

feature be taken as reference i.e. FCNN+GBML. The time and amount are the main class

parameters there is no null value present in the dataset. As per proposed algorithm results

284315 class zero i.e normal transactions are identified and 492 class 1 i.e fraud transactions

are identified.

4.2. Performance Analysis

DANCON2 and local datasets are being used to identify the transaction is fraud or not. The

entire technique identify the 493 frauds out of 284997, which is a highly balanced that is 0.173

to solve this class FCNN and GBML distribution classes are applied into training and testing

28499

1000

0

5000

10000

15000

20000

25000

30000

Real frauds

FRAUDS ANALYSIS




sets. The entire dataset is split into training sets-2 and independent test-1. The comparison table,

which is shown in below.

Table 3 Attributes

Variable name Description Type

A0, A2, A3, .An,

Transaction features after FCNN “Integer

Time GBML Integer

Amount GBML trees Integer

Class Non fraudulent or fraudulent 0 or 1”

Table 3. Explains about GBML analysis and FCNN truncations fraud predictions, based on

binary digital data computations. In this 0 and 1 represents FRAUD or NO FRAUD

respectively.

Table 4 Comparison of algorithms

SVM Linear regression It is proven that SVM is good choice for solving

variant data structures for fraud detection [22].

Regression LR Various credit card frauds are identified and

compete with traditional methods.

ANN Non-linear Nonlinear regression with ANN provides the

better results compared to linear models.

FCNN-GBML Deep learning and machine

learning

Using FCNN auto encoder predictor and

GBML machine learning classifier can achieves

the fraudulent transactions, these are best fitted

for current technology.

Table 4 Explains about comparison algorithms, with respect to cloud disorders

identifications, in this various models and its functionality is described based performance

analysis. FCNN-GBML model can handle real-time problems like dynamic nature of clouds

and variable environment of software Mis-functionalities. By using deep learning “FCNN” and

machine learning “GBML” is utilized to solves the any type of problems which are explained

clearly in above table

Table 5 confusion matrix.

Sl.No. TP FP TN FN

1. SVM 342 2112 115174 235

2. LR 303 1291 116001 271

3. NN 385 2228 115048 199

4. Non linear 431 3870 113409 152

5. FCNN-GBML 359 1458 115794 256

Table: 4. Explains about True positive rate, false positive rate, True negative rate, false

negative rate of all described methods “SVM, LR, NN, Nonlinear and FCNN-GBML”. In this

proposed method got more improvement compete with current models.



Figure 12 Tp analysis

Figure .12 explains about graphical representations of True positive rate vs negative rate,

in this proposed method” FCNN-GBML” archives more improvement compared to survey

methods.

Table 6 F1 scores

Model F1 score

SVM 0.226

LR 0.270

NN 0.241

Non linear 0.177

FCNN-GBML 0.299

Figure 13 F1-score

Table: 5 and Figure 13 explains about computational achievement of F1- score results, in

this proposed method got 0.299 F1- score. This is best improvement compared to reaming

methods which are discussed.

0 100 200 300 400 500

SVM

LR

NN

Non linear

FCNN-GBML1

23

45

True positive vs negitive

FN TP

0.226

0.270.241

0.177

0.299

F1 SCORE

SVM LR NN Non linear FCNN-GBML




Table 7 Accuracy

Model Accuracy

SVM 0.9812

LR 0.982

NN 0.978

Non linear 0.965

FCNN-GBML 0.998

Figure 14 accuracy analysis

Table 6 and Figure .14 explains about Accuracy discriminations of FCNN-GBML method,

in this 0.998 accuracy is achieved. Nonlinear methods, LR, NN, and SVM attains more

accuracy but improvement is needed, proposed method has achieves more Nemours accuracy.

5. CONCLUSION

In this investigation a real time credit card fraud detection system on cloud computing has been

implemented. It is useful for real time transactions, in this a real life database of credit card

transaction utilized at deep learning (FCNN) auto encoder. The GBML is used to classify the

frauds on clouds with effective manner. In this work the accuracy is 0.98 F1 score, 0.299 and

TP is 359 is achieved. Which is a good improvement. This experiment further useful for real

time cloud based fraud identification system. To monitor any time invariant behavior of clouds

this application is helpful for future technologies.

REFERENCES

[1] J.P.Morgan. "Payments Fraud and Control Survey", PNC Financial Services Group, 2018.

[2] Boston Consulting Group. (2017). Global Payments 2017 – Deepening The Customer

Relationship.

[3] Bose, Indranil, and Radha K. Mahapatra. "Business data mining—a machine

learning perspective." Information and management 39.3 (2001): 211-225.

[4] TUNG, Hui-Hsuan, CHENG, Chiao-Chun, CHEN, Yu-Ying, et al. Binary Classification

and Data Analysis for Modeling Calendar Anomalies in Financial Markets.

In : Cloud Computing and Big Data (CCBD), 2016 7th International Conference

on. IEEE, (2016). p. 116-121.

0.9812

0.982

0.978

0.965

0.998

0.94 0.95 0.96 0.97 0.98 0.99 1 1.01

SVM

LR

NN

NON LINEAR

FCNN-GBML

Accuracy



[5] Jacomo Corbo, Carlo Giovine, and Chris Wigley(2017, April). Applying analytics in financial

institutions fight against fraud. McKinsey Analytics. Retrieved from

https://www.mckinsey.com.

[6] Vapnik, V.: Statistical learning theory. Wiley, New York.(1998)

[7] I. Goodfellow, Y. Bengio, A. Courville.: Deep learning, Cambridge, Massachusetts,

The MIT Press, (2016)

[8] Phyu, Thair Nu. "Survey of classification techniques in data mining." Proceedings

of the International MultiConference of Engineers and Computer Scientists. Vol.

1. (2009).

[9] Bolton, Richard J., and David J. Hand. "Statistical fraud detection: A review."

Statistical science (2002): 235-249.

[10] Zhou, Xun, et al. "A state of the art survey of data mining-based fraud detection

and credit scoring." MATEC Web of Conferences. Vol. 189. EDP Sciences, 2018.

[11] John A. Stankovic. :Misconceptions about real-time computing. IEEE Computer,

21(10), 10-19 (1988).

[12] S. Benson Edwin Raj, A. Annie Portia.: Analysis on Credit Card Fraud Detection

Methods. In: International Conference on Computer, Communication and

Electrical Technology – ICCCET2011, 18th & 19th March, (2011)

[13] Martin, James.: Programming Real-time Computer Systems. Englewood Cliffs,

NJ: Prentice-Hall Inc. p. 4. ISBN 0-13-730507-9. )(1965)

[14] Masoumeh Zareapoor, Seeja.K.R, and M.Afshar.Alam.: Analysis of Credit Card

Fraud Detection Techniques: based on Certain Design Criteria. In: International

Journal of Computer Applications (0975 – 8887) Volume 52– No.3, August (2012)

[15] Bhattacharyya, Siddhartha, et al. "Data mining for credit card fraud: A comparative

study." Decision Support Systems 50.3 (2011).

[16] Ghosh, Sushmito, and Douglas L. Reilly. "Credit card fraud detection with a

neural-network." System Sciences, 1994. Proceedings of the Twenty-Seventh

Hawaii International Conference on. Vol. 3. IEEE, (1994).

[17] Bakar, Nor Mazlina Abu, and Izah Mohd Tahir. "Applying multiple linear regression

and neural network to predict bank performance." International Business

Research 2.4 (2009): 176.

[18] Landi, Alberto, et al. "Artificial neural networks for nonlinear regression and

classification." Intelligent Systems Design and Applications (ISDA), 2010 10th

International Conference on. IEEE, (2010).

[19] Aleskerov, Emin, Bernd Freisleben, and Bharat Rao : Cardwatch: A neural network

based database mining system for credit card fraud detection. Computational Intelligence

for Financial Engineering (CIFEr), 1997., Proceedings of the IEEE/IAFE

1997. IEEE, 1997.

[20] Dorronsoro, Jose R., et al. : Neural fraud detection in credit card operations. IEEE

transactions on neural networks 8.4 (1997).

[21] Seeja, K. R., and Masoumeh Zareapoor. "FraudMiner: A novel credit card fraud

detection model based on frequent itemset mining." The Scientific World Journal

2014 (2014).

[22] Hassibi, Khosrow. "Detecting payment card fraud with neural networks." World

Scientific Book Chapters (2000): 141-157.

[23] ALTMAN, Edward I., MARCO, Giancarlo, et VARETTO, Franco.: Corporate

distress diagnosis: Comparisons using linear discriminant analysis and neural

networks (the Italian experience). Journal of banking & finance, vol. 18, no 3, p.

505-529.(1994)




[24] D.W. Hosmer, S. Lemeshow : Applied Logistic Regression, 2nd Ed, Wiley-

Interscience, (2000).

[25] GOH, King-Shy, CHANG, Edward, et CHENG, Kwang-Ting.: SVM binary classifier

ensembles for image classification. In : Proceedings of the tenth international

conference on Information and knowledge management. ACM. p. 395-402.(2001).

[26] Raju, K., Pilli, S.K., Kumar, G.S.S., Saikumar, K., Jagan, B.O.L.(2019). Implementation of

natural random forest machine learning methods on multi spectral image compression, Journal

of Critical Reviews6(5), pp. 265-273

[27] Wang, Z., Wang, N., Su, X., & Ge, S. (2020). An empirical study on business analytics

affordances enhancing the management of cloud computing data security. International Journal

of Information Management, 50, 387-394.

[28] Giri, S., & Shakya, S. (2019). Cloud Computing and Data Security Challenges: A Nepal

Case. International Journal of Engineering Trends and Technology, 67 (3), 146, 150.

[29] Garg, S., Singh, A., Kaur, K., Aujla, G. S., Batra, S., Kumar, N., & Obaidat, M. S. (2019). Edge

computing-based security framework for big data analytics in VANETs. IEEE Network, 33(2),

72-81.

[30] Nayyar, A. (2019). Handbook of Cloud Computing: Basic to Advance research on the concepts

and design of Cloud Computing. BPB Publications.

an implementation of real and accurate cloud based … · implementation of real and accurate cloud...

Documents